AI Is Worse Than The Dot Com Bubble: Part Two

0:00This is an iHeart Podcast. Guaranteed human. Run a business and not thinking about podcasting? Think again. More Americans listen to podcasts, then add supported streaming music from Spotify and Pandora. And as the number one podcaster, iHeart's twice as large as the next two combined. Learn how podcasting can help your business. Call 844-844-IHEART. Call Zone Media. I'm Ed Zetron, and this is Better Offline.

0:39that's right folks we're back for the second part of our series on the dot-com bubble and why i believe the ai bubble could be much much worse while the dot-com bubble was a mixture of dodgy venture capital deals and websites that could never turn a profit combined with a global mania around the interconnectivity of high-speed internet companies the ai bubble is one company selling expensive AI GPUs, a bunch of companies building data centers to put them in, and a bunch of companies building shit that runs on GPUs that only loses money and that customers kind of fucking hate. I should also note that a very important part of the story is venture capital's lack of returns in the last few years, something I covered in the in-shit-to-financial crisis last week, specifically in parts two and three.

1:19In simple terms, AI startups now make up more than half of venture investment, and I believe that most of these startups will die because of their horrible margins, no path to profitability, and products that people really don't want to pay for at scale. And when they die, they will leave venture capitalists stranded with tons of dead equity at a time when they already have trouble generating returns and thus raising money from their limited partners. The result I worry will be gruesome. Venture capitalists make their money through the fees they generate, which are based on the value of their investments and the returns they give their investors, which don't seem to be happening.

1:54What do you think happens when they can't generate any returns and their investments aren't worth anything? The answer is simple. They won't have any way of raising more capital as their limited partners won't fucking trust them. And to be clear, these R-swipes have cocked it up many years at a time. Look at crypto, look at NFTs, look at AR, VR, metaverse, all of that. And all of this ridiculousness has happened because of the ridiculous, ridiculous myths of AI, like the fictitious, insatiable demand for AI compute and the made-up decline in the price of intelligence which by the way that last one i've talked about it i swear to god i mentioned it in the garden i mentioned everywhere it may be cheaper to pay for the tokens but you use more of them so it's more expensive in the aggregate on top of that it doesn't mean that the price of intelligence for the model providers is going down jesus christ I'm so tired of making this point.

2:48But in both cases, these assumptions are convincing investors that it's time to invest in data centers that only lose money because you assume that the demand will be there, or in AI startups that only lose money because you'll assume they magically stop losing money somehow. And it turns out we do actually have a historical comparison. The mania of the dot-com bubble was based on a misunderstanding of the scale of the internet at the time rather than its actual potential. Hundreds of billions of dollars were invested based on flimsy logic. To quote researcher Justin Kohler, This continental rewiring was also justified by another powerful myth, that internet traffic was doubling every 90 days.

3:27This claim spread through analyst reports, earning calls, and investor presentations like a particularly virulent meme. If true, it meant that demand was growing exponentially, far outpacing any conceivable supply, and that every new trench of fiber would soon pay for itself many times over. And I pause here to go, oh God, they're doing it again. Back to the quote. But the mathematics were fiction. Network researchers like Andrew Adlisko at AT &T, looking at actual traffic data, found that US backbone traffic was doubling roughly once a year. Rapid growth, certainly, but nowhere near the purported 90-day cycle.

4:04Meanwhile, advances in fiber technology were making each strand exponentially more powerful. dense wavelength division multiplexing allowed dozens of signals to travel simultaneously down the same line at different wavelengths of light like multiple conversations happening in different colors while demand doubled annually supply expanded tenfold or more carriers buried the discrepancy under layers of creative accounting that would have impressed medieval alchemists first of all justin if you hear this i fucking love that that was very fun second of all to To quote Twin Peaks, it's happening again.

4:39Mania had taken hold based on very flimsy logic. Economically speaking, this meant that telecoms companies, server hardware companies, ISPs, construction firms, optical cable providers, wireless technology companies, and basically anybody related to the business of providing internet access in any way saw a massive influx of business to build capacity that didn't need to be built yet. If you were in the business of selling services to get people online, you were high on the hog. You know, kind of like selling high bandwidth RAM. Similarly, one could get a startup funded if you had a website or even take it public.

5:12One could raise debt to build a nascent ISP or a fiber network. The money was flowing because people weren't really being thoughtful about it. And as far as the dot-com part of the dot-com bubble, the unsustainable websites, the problem wasn't so much the industry but the businesses themselves, which were hyped and dumped onto the public markets with little regard for their long-term health. the problem here is relatively simple these were bad companies that people ignored the issues with because of and i quote the power of the internet and how it would somehow save them which which it didn't obviously had they been kept private and died in the dark i don't think these companies would have had the same reputation i don't think we give a fuck about pets.com i also don't think they were really an accurate comparison to anything happening today while the valuation's ridiculous The globe's market cap was$1.840 million.

6:04The scale of destruction caused by dot-com startups was significantly smaller, even in today's money. The economics were bad, but not anywhere near as bad. For the first nine months of 1998, the globe made$2.7 million in revenue and lost$11.5 million, largely due to trying to move into multiple different business lines at once, like voice over IP. And to be clear, this company made money selling ads and also buying random companies buying random companies was the thing that happened during the dot-com boom everybody fucking loved buying companies you just bought random there was uh i think like excite bought at home like there was a the at &t sorry not the aol time warner merger so many stupid mergers so little time nevertheless the economics of this shit show were quite complex you had companies raising money to do any website they could think of companies raising money to lay fiber companies raising money to found ISPs, all of which had multifaceted layers of physical and digital infrastructure that were quite...

7:04unbuilt, I think is the term. It's tempting, yet incorrect to say the thing about AI. The similarity everybody points to is that people doubted the internet at the time, and people really need to remember their fucking history. In 2000, only 52 % of Americans were using the internet, and by 2003, their number had only increased to 61%. Per the World Bank, in 2005, only 16 % of the world used the internet, and in 2024, that number had increased to 71%. Yet the real difference is the access to high-speed internet. When the internet was connected via a 56k modem, access was at times charged by the minute, and even if it was unlimited, it was always much, much slower.

7:45While we're used to connecting at speeds that make using a web-based app near indistinguishable from using one on a computer, back in 2000, 2001 or 2002, the average US internet speed was at best 400 kilobits per second or roughly 50 kilobytes per second compared to the average US internet speed today of over 200 megabits per second or 25 megabytes a second. In simpler terms, and the younger members of the audience won't understand this, a website took time to load in a way that feels almost impossible to conceive if you didn't experience it at the time. You had to make a commitment to go to a website.

8:19It wasn't like you'd browse multiple tabs and fuck around in different windows. You sat there and you waited a little bit. Sometimes it came up quicker than another. In fact, websites like Google were quite popular because they were very clean. And the reason that them being clean wasn't usability, it was the fact it loaded quickly, which I guess would be usability. Either way, we've also had dramatic improvements in web design and accessibility, the advent of mobile browsing, and the proliferation of widespread mobile and desktop internet access. In the 2000s, we were at the very early days of e-commerce, and the weird irony of the dot-com bubble is that it was actually pretty useful to lay millions of miles of fiber optic cable.

8:59This is, in no way, shape or form remotely comparable to large language models, GPUs, or any nebulous VC spunk around generative AI.

9:40Thank you.

9:47844-844-IHEART to get started. That's 844-844-IHEART.

9:55Global internet access has never been higher or cheaper, and for the most part, billions of people can access the connection fast enough to use generative AI. There is very little stopping anyone from using an LLM. ChatGPT is free. ChatGPT's cheaper Go subscription has now spread to the entire world. When I originally wrote this section, it was originally just in the global south but now it's everywhere gemini is free perplexity is free and metas llm is free where the dot-com bubble was made up of stupid businesses and the lack of fundamental infrastructure to give most people the opportunity to access a reliable internet experience basically anybody can get reliable access to generative ai anyone claiming this is just like the early days of the internet is a fucking liar or a fucking moron llms have now spread to every nook and cranny of the internet anybody can use one anybody can experience the so-called power of ai users are not sitting frothing at the mouth unable to access chat gpt due to a lack of infrastructure nor is anybody saying oh man i can't access claw because i don't have a local data center they might be doing it because there's a fucking rate limit because anthropic can't afford to run their services but that's not what this is about edward experiences are not worse because these companies have a lack of access to infrastructure or capital.

11:13They're worse because the underlying technology of transformer-based models is inherently limited, and in turn, any company connected to these models is limited along with them. Then we get to the economics of the AI bubble, and things begin to get more worrying. While the dot-com bubble rested on the back of companies like Lucent, Cisco, Nortel, Worldcom, Enron, and others, the AI bubble rests fundamentally on one company, NVIDIA, and to a lesser extent, the valuations of the remainder of the Magnificent Seven, Microsoft, Amazon, Google, Meta, Apple, and Tesla. Four of those companies, Amazon, Microsoft, Google, and Meta, through intermediaries I'll get to in a future episode, I think, actually I'll explain in a second, have spent hundreds of billions of dollars on GPUs and their associated infrastructure for reasons that none of them can seem to explain.

12:01as an aside by the way i will get to this in an episode i had to cut it from the script just for length it's already quite long um there is a weird thing going on where microsoft google meta amazon they don't buy their gpus directly from nvidia they get them through various taiwanese uh holding companies like fox holding companies the wrong word manufacturers of server hardware and such called like honhai who is voxcon wish tron quantum computing they order through taiwan and then those servers are put together and shipped to their data centers this allows them to hide how many gpus they're buying from their investors because guess what it's quite a lot anyway uh when this all collapses we're also going to see a market contagion that goes to taiwan because all those taiwanese companies are booking revenue from selling these fucking servers anyway lots of fun there but let's keep going now nvidia's revenue is also predominantly 88 in its data center segment and its customers are those who can afford at the very least 50 to 100 gpus retailing at 400 grand or more for a pod of eight of them and you require tens of thousands of dollars of networking gear to go with them to make them turn on the customers of those renting those gpus are either ai labs training or running inference for models and their customers are ai startups the problem isn't so much that nobody can afford a GPU, but that you can't get very far with just one.

13:27You have to buy so many of them, you need to build a big data center around them, you need to get power to that data center, and then you have the massive environmental concerns of, well, running all that power. This naturally means that there are really only two customers who can afford these chips at scale. The magnificent seven who have all now begun to take on debt after previously financing their GPU purchases with cash flow, and companies that raise debt with companies, meaning anybody who wants to build a data center an oracle who had negative 13 billion dollars in cash flow last quarter and is steeped in debt to the point that bondholders are suing them we also have no idea if the economics of renting gpus actually makes sense and based on everything i've found i'm not sure anybody renting them can ever make a profit due to either or both the upfront cost and debt necessary to pay it and the power intensive nature of providing ai compute it is fundamentally insane and that we don't know for sure it's so crazy how do we not know how the fuck do we not know that this is crazy it's what we do know is that the only company making any kind of profit during the ai bubble appears to be nvidia or companies selling ram microsoft google meta and amazon refuse to share their actual ai revenues and because people have the brains of dogs They have conflated revenue growth from hyperscalers' already existing segments like software and advertising with growth created by AI.

14:52In the dot-com bubble, one could at the very least point to where a company was making revenue, even if the answer was handing a dollar to somebody and getting handed the dollar back. People bought and installed physical infrastructure, and that infrastructure was, albeit at a much level scale than the build-out anticipated, paid for by the associated services. Companies got greedy, rushed to expand in a way that was unnecessary, took on ruinous debt and suffered the consequences. This isn't what's happening in the AI bubble. Consumers have no problem getting exposure to AI. In fact, AI is breaking into every single device and app that we have like an angry pervert with a knife.

15:28While WorldCom wannabes like OpenAI and Anthropic are whining about not having enough compute, it's very clear they've got more than enough to burp out a new model every few months or drop copyright infringement machines on millions of people at a moment's notice the post bubble overbuild of fiber lift thousands of miles of dark i.e not connected to anything cabling that took years to light but doing so had an obvious business use case connecting people to the internet and didn't require an entire fucking data center and masses of power to do so to make matters worse as i've hinted at the depreciation of these gpus is utterly brutal per paul kudroski and i quote we are in a historically anomalous moment.

16:08Regardless of what one thinks about the merits of AI or explosive data center expansion, the scale and pace of capital deployment into a rapidly depreciating technology is remarkable. These are not railroads. We aren't building century-long infrastructure. AI data centers are short-lived, asset-intensive facilities riding declining cost technology curves, requiring frequent hardware replacement to preserve margins. Let me put it a little simpler. imagine if all of that fiber was useless in five or six years at best what if all of that fiber could only be used to access a small subset of websites what if all of that fiber required such massive amounts of power that it threatened rolling blackouts of the east coast of america that is the scale of the apocalypse i am talking about and i am worried that people are not taking the problem more seriously the demand for nvidia chips is fueled by hype and that hype has caused this company and to a lesser extent the magnificent seven to become a load-bearing part of the american stock market an analysis from portfolio manager donkey wong from january 2025 found that the magnificent seven stocks accounted for 47.87 of the russell 1000 indexes returns in 2024 and that's an index fund of the thousand highest ranked stocks on the footsie russell's index in really simple terms without the mostly vibes driven nature of the magnificent seven's growth as now if this is based on anyone's actual revenues, the US stock market would be in incredibly rough shape.

17:38Except unlike the dot-com bubble, most of these companies have taken on incredibly large amounts of GPUs, debt, finance and operating leases, and data centers full of GPUs that can't be used for really much of anything else. And because GPUs are guaranteed to depreciate, each and every one of them will, without fail, have to write down the value of upwards of$100 billion of investments in the future, as these things are eventually facing the recoverability test, which is when there's a huge crash within any market sector and you have to look at your assets and say, shit, will these actually generate enough money?

18:11And this is going to happen whether or not the AI bubble bursts, as NVIDIA is on a yearly cycle of upgrades on their GPUs. Every single year, every single GPU investment loses value. And to make matters worse, it takes fucking years to install these things. So by the time they're there, you're way in the past. Even if the AI bubble doesn't burst, it's gonna. The US stock market has an unhealthy relationship with NVIDIA, which by this time next year will have to make over$90 billion a quarter to keep up with its ridiculous 50 % year-over-year growth. And by 2028, NVIDIA will, to keep its ridiculous valuation, have to be making more than Apple, which makes about$416 billion a year in revenue.

18:52In fact, from my calculations, NVIDIA will have to be making$500 to$600 billion, which puts it in the realm of Walmart. It can't happen. It can't happen. It can't happen. And to do that, NVIDIA's customers will continue having to be able to afford these GPUs which, as I've established, are being paid out of debt because AI services do not make a profit. Even if AI services take off and are useful in a way they've never even remotely hinted at being, it is inevitable that the debt and cash necessary to keep buying NVIDIA GPUs runs out. And more than likely, the revenues of the Magnificent 7 will stumble in growth before then, as it becomes obvious that those GPUs are not providing any meaningful revenue growth.

19:35The result, I fear, is that the American stock market takes a shit the size of Iowa. And due to the unique way that the tech industry functions, the contagion will be global. I'll catch you tomorrow for part three. I don't have a rosy or funny app of it. Every time I think of this stuff, I feel very, very sad. Anyway, very optimistic. Peace. I'll catch you tomorrow.

20:22BetterOffline.com or visit BetterOffline.com to find more podcast links and of course my newsletter. I also really recommend you go to chat.wheresyoured.at to visit the discord and go to r slash BetterOffline to check out our Reddit. Thank you so much for listening. Better Offline is a production of Cool Zone Media. For more from Cool Zone Media, visit our website, coolzonemedia.com. Or check us out on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.

21:10This is an iHeart Podcast. Guaranteed human.